Repository navigation
Conversation
…line - Add install/prepare/typical_deployment_scenarios.mdx. It describes a Single Cluster and a Multi-Cluster platform on Alauda OS, with an optional in-cluster Ceph setup for workload clusters: per-node CPU, memory, and disks of each node pool, node pool and control plane endpoint choices, feature and storage selection, the plugins placed on infra nodes, and the installation steps in order. - Link the new page from the deployment architecture section of Plan. - Add a small three-node baseline to the Ceph aggregate planning budget and a 100 GB persistent file system per MON to the storage device planning. - Update llms.txt and llmstxt-state.json for the three pages. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Deploying alauda-container-platform with
|
| Latest commit: |
dd33a44
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://4df772dd.alauda-container-platform.pages.dev |
| Branch Preview URL: | https://docs-install-typical-deploym.alauda-container-platform.pages.dev |
…tral registry for multi-cluster - Rename the page to Typical Installation Scenarios, because Deployment is a Kubernetes resource name, and update the link from Plan and the llms files. - Describe the storage choices as the storage that ACP provides, and state that Alauda Storage can also serve as the main storage of a large platform. - Transpose the disk tables so that each row is a disk with its purpose and each column is a node pool, and split the multi-cluster disk table into the global cluster, size A, and size B. - For a Multi-Cluster platform, use a central registry: the existing one, or Harbor on the global cluster or in a dedicated DevOps cluster. Remove the ACP Registry from the workload cluster plugins and the Ceph setup. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nd resize the typical scenarios - Evaluating Resources for Workload Cluster: add a Single Cluster table for a global cluster that also runs the workloads (24, 120, and 254 worker nodes: 8/16, 16/32, and 32/128 cores/GB), with the reasons the workload cluster table does not apply. - Evaluating Resources for Global Cluster: state that the page covers only the global cluster of a Multi-Cluster deployment. - Plan: link the Single Cluster row to the new table instead of a fixed recommendation. - Typical Installation Scenarios: resize the node pools to the product capacity tiers (single-cluster control plane 8/16, global infra 12/24, size A control plane 4/16, size B ingress 4/8, monitor and log 8/16, Ceph storage 8/24), and update the sizing notes. - Update llms.txt and llmstxt-state.json for the four pages.
… Gateway - Disk Configuration: organize Storage Capacity by environment and node role. IaaS providers use a system disk plus separate etcd, containerd, kubelet, and /var/cpaas disks, with only /var/cpaas persistent; Bare Metal uses one system disk. Add path sizes by role, the raw disk for TopoLVM, Bare Metal system disk sizes, and a summary by node role. - Typical Installation Scenarios: list disks per node for IaaS providers and Bare Metal and link Disk Configuration for the reasons; use 1 TB and 512 GB raw disks. - Use Envoy Gateway (Gateway API) for application traffic, rename the size B ingress pool to gateway, and place envoy-proxy on infra nodes. - Allocate the control plane and gateway VIPs from an enterprise load balancer when one exists. On Huawei Cloud Stack, run envoy-proxy with the host network behind the HCS ELB instead of MetalLB. - Recommend 4 Ceph storage nodes on Bare Metal. - Tasks for Envoy Gateway: add anchors for the envoy-proxy scheduling and hostNetwork sections. - Update llms.txt and llmstxt-state.json.
…link the DevOps planning - Evaluating Resources for Global Cluster: add Global Cluster as a Workload Cluster (moved from the workload page, anchor kept), and scope Node Sizing to Multi-Cluster deployments. Evaluating Resources for Workload Cluster points to the new section; Plan and Typical Installation Scenarios link it. - Ceph planning: state that the small baseline is for three nodes and the other baselines for four nodes, and recommend four storage nodes on Bare Metal. - Disk Configuration: merge the size columns of the path table into one "Size and notes" column, name the node role columns clearly, and state that Harbor and the DevOps tools keep their data in Ceph and need no TopoLVM disk, while DevOps cluster infra nodes use a 1 TB TopoLVM disk. - Typical Installation Scenarios: list the harbor and storage pools for Harbor on the global cluster, and link the DevOps planning page. - Update llms.txt and llmstxt-state.json.
The at-a-glance section is being removed from the DevOps planning page. Each case section there lists its own machines, disks, and totals, so Central Registry links only the case sections.
…recovery By default, add the harbor and storage pools to the primary global cluster only. If Harbor needs a standby, prepare the same pools on the standby.
| ::: | ||
| | Environment | Disks of each node | | ||
| |---|---| | ||
| | Alauda OS on Huawei DCS, Huawei Cloud Stack, or VMware vSphere | A system disk, plus a separate disk for each of `/var/lib/etcd` (control plane nodes only), `/var/lib/containerd`, `/var/lib/kubelet`, and `/var/cpaas`. Declare only `/var/cpaas` as a persistent disk, so that it is kept when the provider replaces the node. Declare the other three as standard disks. | |
There was a problem hiding this comment.
这句话会不会有歧义?用户误以为只需要一个独立的磁盘就行,其实是每个路径都需要一个独立的磁盘。
并且排版优化下,现在读的很累
There was a problem hiding this comment.
已改。Disk Layout by Environment 按环境分成三个小节:IaaS 用列表逐块列出系统盘、etcd 盘(只有控制面节点)、containerd 盘、kubelet 盘、/var/cpaas 盘(声明为持久盘),并写明每个路径各一块独立的盘;Bare Metal 是一块系统盘;non-Alauda OS 同 Bare Metal,用 /cpaas 替代 /var/cpaas。
|
|
||
| ### Path Sizes \{#path_sizes} | ||
|
|
||
| | Path | Used for | Size and notes | |
There was a problem hiding this comment.
已改。路径表每个单元格只写一句话,例如「At least 10 GB; 20 GB recommended.」;高 IO 盘、80% 利用率、日志存储节点指哪些、etcd 备份位置这类说明都挪到表下的列表里。
- sites.yaml: set the devops version to 4.4. - Upgrade the Builds Operator: link the Tekton v3 to v4 migration guide at upgrade/migrating-tekton-v3-to-v4.html. The previous 4.0/upgrade/ path does not exist on any release of the DevOps documentation.
|
|
||
| The logging size covers about 5,000 log entries per second with 7 days of retention. For more logs, add capacity to the TopoLVM volume group. On Bare Metal, raw disks usually come in sizes such as 512 GB and 1 TB; select the smallest size that covers the need. | ||
|
|
||
| Harbor and the DevOps tools need no TopoLVM disk. Harbor, GitLab, SonarQube, Nexus, and their PostgreSQL and Redis instances keep their data in Ceph (RBD and RGW). For the OSD disks of the storage nodes, see <ExternalSiteLink name="devops" href="/install/planning-devops-cluster.html#harbor-on-global" children="Harbor on the global Cluster" /> and <ExternalSiteLink name="devops" href="/install/planning-devops-cluster.html#devops-cluster" children="Dedicated DevOps Cluster" />. The infra nodes of a DevOps cluster run monitoring and logging, like the infra nodes of a workload cluster, and use a 1 TB TopoLVM disk. |
There was a problem hiding this comment.
这句话不太准确。并不是不需要,只是如果 DevOps tools 不选择用 TopoLVM 作为存储,才不需要。
如果在一些极端交付场景中使用 TopoLVM 作为存储,也是需要评估磁盘大小的。但这个不用写在文档里,也不要因为这个东西去改 DevOps 那个 PR。只是说,你当前这句话不能写得太绝对。
There was a problem hiding this comment.
已改,不再写成绝对的「不需要」。现在写的是:按 DevOps 规划页的方案,Harbor、GitLab、SonarQube、Nexus 及其 PostgreSQL、Redis 的数据都在 Ceph 上,因此不需要 TopoLVM 盘。没有提极端场景,也没有改 DevOps 那个 PR。
|
|
||
| Harbor and the DevOps tools need no TopoLVM disk. Harbor, GitLab, SonarQube, Nexus, and their PostgreSQL and Redis instances keep their data in Ceph (RBD and RGW). For the OSD disks of the storage nodes, see <ExternalSiteLink name="devops" href="/install/planning-devops-cluster.html#harbor-on-global" children="Harbor on the global Cluster" /> and <ExternalSiteLink name="devops" href="/install/planning-devops-cluster.html#devops-cluster" children="Dedicated DevOps Cluster" />. The infra nodes of a DevOps cluster run monitoring and logging, like the infra nodes of a workload cluster, and use a 1 TB TopoLVM disk. | ||
|
|
||
| ### System Disk on Bare Metal \{#bare_metal_system_disk} |
There was a problem hiding this comment.
这里的内容和下面的 Disks by Node Role 是不是有冲突啊?或者说只需要留下面那个,然后这里可以删掉?
There was a problem hiding this comment.
对,两处是重复的。已删掉「System Disk on Bare Metal」一节,Bare Metal 系统盘的最低值和推荐值并进「Disks by Node Role」表,各占一行;monitor、gateway、storage 节点在表下说明按 Application worker node 那一列取值。原来的锚点 #bare_metal_system_disk 没有别处引用,已随这一节删除。
|
|
||
| | Disk | `global` control plane node | `global` infra node | Workload control plane node | Workload infra or log node | Application worker node | | ||
| |---|---|---|---|---|---| | ||
| | **Alauda OS on Huawei DCS, Huawei Cloud Stack, or VMware vSphere** | | | | | | |
There was a problem hiding this comment.
Alauda OS on Huawei DCS, Huawei Cloud Stack, or VMware vSphere
这个太长了吧?你要不就简化成 IaaS,或者 On IaaS。你看哪个更合适?可以在表格上面或者下面说明一下,对 On IaaS 的场景做举例。
There was a problem hiding this comment.
已改为「IaaS」,本节开头写了一句:IaaS 指 Alauda OS 运行在 Huawei DCS、Huawei Cloud Stack、VMware vSphere 上。表里的环境行现在是 IaaS、Bare Metal、All environments,场景页的磁盘表和 Ceph 台数说明也统一成同样的写法。
| | Single Node | You need a test or proof-of-concept environment. | Do not use Single Node for production. On Alauda OS, a single control plane node is supported for test environments, and a cluster with a single control plane node cannot be upgraded. See `controlPlane.replicas` in [Installation Reference](../alauda_os/reference.mdx#install-config). | | ||
|
|
||
| For the nodes that each architecture needs, see [Nodes of the `global` Cluster](#global-nodes). | ||
| For the nodes that each architecture needs, see [Nodes of the `global` Cluster](#global-nodes). For the node pools, machine sizes, and disks of a typical Single Cluster or Multi-Cluster platform on Alauda OS, see [Typical Installation Scenarios](./typical_installation_scenarios.mdx). |
There was a problem hiding this comment.
这个排版也有问题啊,是不是得换个行啊?
There was a problem hiding this comment.
已改。架构表下面那句长句拆成了三条:一条写各架构的控制面规格并链接 evaluating_global 的对应小节,一条写 Alauda OS 上的标准节点池、机器、磁盘和插件并链接 Typical Installation Scenarios,一条是 Single Node 的 controlPlane.replicas 说明。
| - **Self-built VIP**: needs no external device. | ||
| - **An enterprise load balancer**: use it if one is available. | ||
|
|
||
| When an enterprise load balancer is available, allocate one VIP on it for the control plane endpoint. The VIP of the application gateway can also come from the same load balancer. On Huawei Cloud Stack, use the ELB that the provider creates. For the modes that each provider supports and what to prepare, see [Plan the Control Plane Endpoint](../alauda_os/prerequisites.mdx#control-plane-endpoint). |
There was a problem hiding this comment.
已改成分条,只写选型:Self-built VIP;企业负载均衡,从上面分一个 VIP 给控制面端点,应用 Gateway 的 VIP 也可以从同一个负载均衡上分;Huawei Cloud Stack 用 provider 创建的 ELB。要准备什么链接到 Plan 的决策表和 Plan the Control Plane Endpoint。
|
|
||
| #### Control Plane Endpoint | ||
|
|
||
| The `global` cluster and every workload cluster need a control plane endpoint. The choices are the same as for a Single Cluster: a Self-built VIP or an enterprise load balancer. When an enterprise load balancer is available, allocate one VIP on it for the control plane endpoint. The VIP of the application gateway can also come from the same load balancer. On Huawei Cloud Stack, use the ELB that the provider creates. See [Plan the Control Plane Endpoint](../alauda_os/prerequisites.mdx#control-plane-endpoint). |
There was a problem hiding this comment.
多集群这一处也改成同样的分条写法:Self-built VIP、企业负载均衡(给每个控制面端点分一个 VIP,应用 Gateway 的 VIP 也可以从同一个负载均衡上分)、Huawei Cloud Stack 的 ELB。要准备什么同样链接到 Plan 的决策表和 provider 页。
| | --- | --- | --- | | ||
| | Monitoring | VictoriaMetrics, one per cluster | TopoLVM | | ||
| | Logging | ClickHouse, one per cluster | TopoLVM | | ||
| | Application gateway (workload clusters) | Envoy Gateway (Gateway API), which uses the Gateway API rather than the Ingress API | — | |
There was a problem hiding this comment.
, which uses the Gateway API rather than the Ingress API
这句是多余的。我之前的 comment 只是给你解释,并不是让你写上去。
There was a problem hiding this comment.
已删掉,单集群和多集群两处都只保留「Envoy Gateway (Gateway API)」。
- Plan: reorganize into Choose a Deployment Architecture, Prepare Resources (machines and network), Confirm Install-Time Decisions (one table that absorbs the addresses, certificate, packages, and registry content), and Next Steps (one entry page per path). Remove the #addresses and #packages-and-registry anchors, which no other page links, and point the two prerequisites pages to the new decisions table. - Typical Installation Scenarios: state that it follows Plan, list the control plane endpoint choices, name the environments IaaS and Bare Metal, move provider links into a Provider Pages table, and keep table cells to one sentence. - Disk Configuration: list the disks of each environment, keep path table cells to one sentence with notes below, merge the Bare Metal system disk sizes into Disks by Node Role, and qualify the DevOps statement on TopoLVM. - Update llms.txt and llmstxt-state.json.
…ipeline pool optional Follow the updated DevOps planning page: the harbor pool on the global cluster is 3 nodes with 12 cores and 24 GiB, and a dedicated DevOps cluster needs a pipeline pool only when 3 or more PipelineRuns run at the same time.
| | --- | --- | --- | | ||
| | Monitoring | VictoriaMetrics | TopoLVM | | ||
| | Logging | ClickHouse | TopoLVM | | ||
| | Application gateway | Envoy Gateway (Gateway API), which uses the Gateway API rather than the Ingress API | — | |
There was a problem hiding this comment.
, which uses the Gateway API rather than the Ingress API
这句是多余的。我之前的 comment 只是给你解释,并不是让你写上去。
There was a problem hiding this comment.
已在 ffb05ff 删除,单集群和多集群两处都只保留「Envoy Gateway (Gateway API)」。
| | Monitoring | VictoriaMetrics | TopoLVM | | ||
| | Logging | ClickHouse | TopoLVM | | ||
| | Application gateway | Envoy Gateway (Gateway API), which uses the Gateway API rather than the Ingress API | — | | ||
| | Application gateway VIP | An enterprise load balancer if available; otherwise MetalLB. On Huawei Cloud Stack, run `envoy-proxy` with the host network and put the HCS ELB in front of the gateway nodes, without MetalLB; see [How To Use `hostNetwork` In `envoy-proxy`](../../networking/how_to/tasks_for_envoy_gateway.mdx#envoy_proxy_hostnetwork). | — | |
There was a problem hiding this comment.
On Huawei Cloud Stack, run envoy-proxy with the host network and put the HCS ELB in front of the gateway nodes, without MetalLB; see How To Use hostNetwork In envoy-proxy.
这句话是不是从表格里面挪出来,放在表格下面更好?感觉放在表格里面,整个表格因为这一句话变得非常臃肿,完全看得不是很清楚。
There was a problem hiding this comment.
这句在 ffb05ff 已经挪出表格了:单集群里,它是功能选型表下面单独的一段;多集群里,它是功能选型表下列表的第一条。表格里现在只保留「An enterprise load balancer if available; otherwise MetalLB」。
| | Persistent disk for `/var/lib/rook` | 100 | | ||
|
|
||
| - With 3 replicas, usable capacity is about one third of the raw OSD capacity. Add OSD disks for more capacity. | ||
| - On Bare Metal, use 4 storage nodes. A replacement for a failed physical server can take a long time to arrive. With only 3 nodes, Ceph then runs degraded on 2 nodes for that whole time, and maintenance operations and upgrades are blocked. With 4 nodes, 3 nodes remain after one failure, and Ceph can restore all 3 replicas. |
There was a problem hiding this comment.
这句话放的位置不对,应该放在"Sizes are per node, in GB. For what each disk holds and why it "的上面
There was a problem hiding this comment.
已改(4109a587)。「Spread the storage nodes across failure domains」和「Bare Metal 上用 4 台 storage 节点」这两条,已经从磁盘表下面挪到 storage 机器表正下方、「Sizes are per node, in GB…」之前;磁盘表下只留可用容量、/var/lib/rook 持久盘和 Ceph 资源预算的说明。
…e Ceph node notes - Single Cluster: 24 worker nodes and 500 namespaces, the first tier of Global Cluster as a Workload Cluster. - Workload size A: 24 worker nodes and 500 namespaces; size B: 120 worker nodes and 1000 namespaces, the first two tiers of Control Plane Node Sizing. - Multi-Cluster with Ceph: place the failure-domain and Bare Metal node count notes directly below the storage machine table.
| | `/var/lib/containerd` | Container images and container writable layers | At least 100 GB; 150 GB recommended. | | ||
| | `/var/lib/kubelet` | Pod ephemeral data, such as `emptyDir` volumes | At least 100 GB; 200 GB recommended on `global` cluster nodes and 100 GB on workload cluster nodes. | | ||
| | `/var/cpaas` on Alauda OS<br></br>`/cpaas` on a non-Alauda OS | Platform data: etcd local backups, the log collection buffer, and audit data | At least 100 GB on control plane nodes and 50 GB on other nodes; 200 GB recommended on log storage nodes. | | ||
| | A location on the machine that runs the installer | The extracted Core Package and Extension packages | At least 120 GB; 250 GB recommended. | |
There was a problem hiding this comment.
这里是 120 和 250 ,为什么 install/alauda_os/prerequisites#bootstrap-host-requirements 里面是 300 GB ?并且明确要求是 /var/cpaas 目录?磁盘到底是要 300 还是 250?
另外我觉得 bootstrap 不应该在 configure/scalability/disk_configuration 这个文档中。bootstrap 在 bootstrap-host-requirements 说清楚就行。
还有,如果用户打算将 bootstrap 也用作 install/prepare/mirror-registry 的机器,在哪里可以知道这个信息,并且怎么知道这个机器要准备多少 cpu 和内存? 4c 8g 300 gb 够吗?
There was a problem hiding this comment.
两部分都已处理。disk_configuration 里「A location on the machine that runs the installer」那一行和表下说明已在 39a4fd7 删除:Alauda OS 的 bootstrap 只在 Bootstrap Node 的 Hardware 小节说明,non-Alauda OS 首个控制面节点要额外留的 120 GB(推荐 250 GB)挪到了 non-Alauda OS prerequisites 新增的 First Control Plane Node 小节。bootstrap 兼作 mirror-registry 的说明在 75a05b8 写进 Alauda OS prerequisites 的新小节「Using the Bootstrap Node as mirror-registry」(#bootstrap-host-mirror-registry):磁盘至少 500 GB(300 GB 给安装,200 GB 给 mirror-registry 数据目录,建议用 LVM),CPU 和内存最低 4c/8G、推荐 8c/16G,装完后不能回收。场景页 bootstrap 那一行只写「可兼作平台镜像仓库」并链接到这一小节,mirror-registry.mdx 没有改。
| - `/var/cpaas` on control plane nodes holds etcd local backups in `/var/cpaas/backup`; see [etcd Backup and Restore](../backup/etcd.mdx). | ||
| - Installer location: the actual size depends on the Extensions that you install. Plan more space if you expect to add components later. | ||
|
|
||
| ### Raw Disk for TopoLVM \{#topolvm_raw_disk} |
There was a problem hiding this comment.
如果用户不是 Bare Metal ,是 Iaas 场景,也需要这么大磁盘?是否不需要这么多?并且如果 Iaas 磁盘扩容是比较方便的,不像 Bare Metal 还要去单独采购。
我觉得这里是不是可以说明下 Iaas 下 topolvm 建议多大磁盘?
There was a problem hiding this comment.
已按 IaaS 磁盘可以扩容来写:Disk Configuration 的 TopoLVM 裸盘表分 IaaS 和 Bare Metal 两列,IaaS 为 750 GB(日志 + 监控 + 镜像仓库)、650 GB(日志 + 监控)、600 GB(只有日志)、100 GB(只有监控),并注明以后可以扩容;Bare Metal 仍是 1 TB / 512 GB。单集群、多集群页的磁盘表已同步。IaaS 裸盘能否扩容已在 PR 描述里列为待确认,因为 Fixed at Creation 目前写的是持久盘大小建好后不能改。
| - The tables give the minimum recommended size of each node. Physical servers are usually much larger, so select servers that are at least this size. | ||
| - The sizes follow the capacity tiers of each product component. After the installation, scale up according to the actual utilization. | ||
| - For ARM nodes, scale the sizes as described in [Machines](./planning.mdx#machines). | ||
| - The scale is a product tier. For a different number of worker nodes or namespaces, select the tier in [Global Cluster as a Workload Cluster](../../configure/scalability/evaluating_global.mdx#single-cluster-control-plane). |
There was a problem hiding this comment.
这句话没看懂,什么是 the scale is a product tier. ?
There was a problem hiding this comment.
已在 39a4fd7 改成直白的说法:单集群写「这些规格对应 Global Cluster as a Workload Cluster 中 24 个 worker、500 个命名空间那一档;规模不同时,在该页选对应的档位」。多集群 workload 的 size A / size B 同样写明分别对应 Control Plane Node Sizing 中 24 / 500 和 120 / 1000 那一档,原来那条 product tier 的说明已删除。
|
|
||
| - **Envoy Gateway**: set the node selector and tolerations of `envoy-proxy` in the `EnvoyProxy` resource. See [How To Set NodeSelector And Tolerations For `envoy-proxy`](../../networking/how_to/tasks_for_envoy_gateway.mdx#envoy_proxy_scheduling). | ||
| - **ClickHouse**: configure the infra toleration and node affinity on the instance. See <ExternalSiteLink name="clickhouse" href="/how_to/deploy_as_logging_storage.html" children="Deploy as ACP logging storage" />. | ||
| - **Kafka**: set the node label and the tolerations when you create the instance. The node label cannot be changed later. See <ExternalSiteLink name="kafka" href="/functions/10-create-instance.html" children="Create Instance" />. |
There was a problem hiding this comment.
ClickHouse 和 Kafka 是不是引用错了,是不是可以合并为引用日志子站点的 https://docs.alauda.io/logging-service/4.4/install/fast-install.html ?
https://docs.alauda.io/logging-service/4.4/install/fast-install.html 有亲和性配置说明吗?
如果有,是否用一句话就行,不用分两行?
为什么这里没有 监控的引用?监控的引用是 https://docs.alauda.io/container_platform/4.5/observability/monitor/install_monitor#acp-monitoring-with-victoriametrics 吧?
并且整个 "Envoy Gateway, ClickHouse, Kafka, and the registry are not scheduled to infra nodes automatically. Configure their placement when you install them:" 这一大段可以删了吗?改为在Installation Steps 列出来会不会更好?
There was a problem hiding this comment.
已删掉这一段和下面的列表,改为在 Installation Steps 的每一步写放置方式。日志(ClickHouse 和 Kafka)合成一步,引用 logging-service 4.4 Fast Install 的 §3.1「Optional — pin pods to a dedicated node pool」,表下写明 install.env 里的亲和性和容忍改成 infra 标签和污点;监控补上 ACP Advanced Monitoring 和它下面「Place VictoriaMetrics Workloads on Infra Nodes」小节。Envoy Gateway 和 Registry 的放置写进各自的步骤;infra 标签值为空,NODE_AFFINITY_VALUE 留空时脚本是否生效,已在 PR 描述里请 reviewer 确认。
There was a problem hiding this comment.
拆分两个页面更好。
分单集群和多集群
比如 prepare 下是 typical scenarios ,typical scenarios 下是单集群和多集群。
There was a problem hiding this comment.
已在 39a4fd7 拆成目录 install/prepare/typical_installation_scenarios/:index.mdx(阅读顺序、两个场景共用的前提、Provider Pages、Other Products)、single_cluster.mdx、multi_cluster.mdx(含 Ceph 子场景)。Plan 里的两处链接已改到新页面,仓库里没有其他页面链接旧页面或旧锚点,lint 0 errors。backport #1443 已同步。
… in the steps - Split Typical Installation Scenarios into a directory with an index, a Single Cluster page, and a Multi-Cluster page that includes Ceph. - State the scenario scales as the matching sizing tier in plain words. - Replace the plugin placement list with a placement note in each installation step, and link the logging and monitoring placement. - Size the TopoLVM raw disk on IaaS by workloads, and keep the Bare Metal sizes. - Move the disk space for the extracted packages to the non-Alauda OS prerequisites.
…stry - Add Using the Bootstrap Node as mirror-registry to the Alauda OS prerequisites: a 500 GB disk, 4 cores and 8 GB minimum, 8 cores and 16 GB recommended, and the machine is kept after the installation. - Link the bootstrap rows of the scenario pages to that section.
Align the harbor pool of the global cluster with the DevOps planning page: 3 nodes of 8 cores and 16 GiB. The storage pool is unchanged.
Remove the harbor and storage pools from the global cluster machine table, and describe the central registry choices without sizes. The machines and disks of Harbor are on the DevOps planning page.
Summary
Adds Typical Installation Scenarios as a directory,
docs/en/install/prepare/typical_installation_scenarios/, with three pages:index.mdx: the reading order (Plan first), the assumptions that both scenarios share, the Provider Pages table, and Other Products.single_cluster.mdx: the Single Cluster scenario.multi_cluster.mdx: the Multi-Cluster scenario, with an optional in-cluster Ceph setup for workload clusters.For each scenario the pages list the per-node CPU, memory, and disks of every node pool, the node pool and control plane endpoint choices, the feature and storage selection, the plugins placed on infra nodes, and the installation steps in order, linking to the existing product pages. A Multi-Cluster platform uses a central registry: the existing one, or Harbor on the
globalcluster or in a dedicated DevOps cluster.Restructures Plan (please review this part closely). Plan is the entry point for every installation; Typical Installation Scenarios follows it for Alauda OS. Plan now has four sections:
#addressesand#packages-and-registryanchors are removed; no other page links them.#prepare-resources,#machines,#global-nodes,#installation-machine, and#networkare kept. The prerequisites pages of both paths now point to the decisions table for packages, the registry, the domain name, and the certificate.Adds an Internal, small three-node baseline (20 logical CPUs / 40 GiB) to the Ceph Aggregate Cluster Planning Budget, and a 100 GB persistent file system per MON to Storage Device Planning. The new page uses them for the Ceph setup.
Sizes the control plane of a Single Cluster separately:
#single-cluster-control-plane), for aglobalcluster that also runs the workloads: 24 / 120 / 254 worker nodes with 500 / 1000 / 4000 namespaces need 8 / 16 / 32 CPU cores and 16 / 32 / 128 GB. Node Sizing on the same page covers the Multi-Cluster case by managed clusters.globalcluster that also runs the workloads to that section.The node pool sizes in Typical Installation Scenarios follow the product capacity tiers. Application traffic uses Envoy Gateway (Gateway API); the size B pool for it is named
gateway. On Huawei Cloud Stack,envoy-proxyuses the host network behind the HCS ELB instead of MetalLB. On Bare Metal, Ceph uses 4 storage nodes.Reorganizes Disk Configuration > Storage Capacity by environment and node role: the disk layout of IaaS providers (system disk plus separate etcd, containerd, kubelet, and persistent
/var/cpaasdisks) and of Bare Metal (one system disk), path sizes by role, the raw disk for TopoLVM, Bare Metal system disk sizes, and a summary by node role. The scenarios page lists only the disks and sizes and links this page for the reasons.Adds anchors to the
envoy-proxyscheduling andhostNetworksections of Tasks for Envoy Gateway.Layout: table cells hold one sentence, paragraphs have at most three sentences, and a sentence has at most two links. In Typical Installation Scenarios, provider links move to a Provider Pages table on the index page, and the Multi-Cluster steps refer to the Single Cluster steps. Environments are named IaaS (Alauda OS on Huawei DCS, Huawei Cloud Stack, or VMware vSphere) and Bare Metal. In Disk Configuration, the Bare Metal system disk sizes are part of Disks by Node Role.
Disk Configuration: the path table uses one Size and notes column; Harbor and the DevOps tools keep their data in Ceph and need no TopoLVM disk; DevOps cluster infra nodes use the TopoLVM sizes of the logging and monitoring row.
Ceph planning: the small baseline is for three nodes and the other internal baselines for four nodes; four storage nodes are recommended on Bare Metal.
Plugin placement: the placement list under Plugins on Infra Nodes is removed. Each installation step now says how to place its component on the infra nodes:
#place-victoriametrics-workloads-on-infra-nodes, with the same ID that the heading already generated).install.envfor the infra label and taint are listed under the steps.Scale wording: the scenario pages state that their sizes match the tier of 24 worker nodes and 500 namespaces (and, for size B, 120 worker nodes and 1000 namespaces) on the sizing pages, and to select the matching tier there for a different scale.
TopoLVM raw disk on IaaS: sized by workloads, because the disk can be expanded later: 750 GB for logging, monitoring, and the registry; 650 GB for logging and monitoring; 600 GB for logging only; 100 GB for monitoring only. Bare Metal keeps 1 TB and 512 GB. The scenario pages use the same sizes.
Installer disk space: the row for the extracted packages is removed from Disk Configuration. On Alauda OS, the bootstrap node is covered by its hardware table in the prerequisites. On a non-Alauda OS, a new First Control Plane Node section of the prerequisites holds the 120 GB minimum and 250 GB recommendation, and Plan links it.
Bootstrap node as mirror-registry: the Alauda OS prerequisites add Using the Bootstrap Node as mirror-registry (
#bootstrap-host-mirror-registry) under the bootstrap hardware: a 500 GB disk (300 GB for the installation and 200 GB for the mirror-registry data directory, on LVM), 4 cores and 8 GB minimum, 8 cores and 16 GB recommended, and the machine is kept after the installation. The scenario pages link it from the bootstrap row and do not repeat the numbers. The mirror-registry page is unchanged.Harbor machines: the
globalcluster machine table no longer lists the harbor and storage pools. Central Registry describes the choices (an existing registry, Harbor on theglobalcluster, or Harbor in a dedicated DevOps cluster) and links the DevOps planning page for the machines and disks, so the sizes are maintained only there.Updates
llms.txtandllmstxt-state.jsonfor the changed pages.The page is the English version of a scenario plan that was reviewed in Chinese in alauda/knowledge#853.
Open items
Depends on docs(appservice): add a storage selection guide for application services #1359. The Application Services row of Other Products should link
appservice/overview/storage_selection.mdx, which docs(appservice): add a storage selection guide for application services #1359 adds. Until docs(appservice): add a storage selection guide for application services #1359 merges, the row links the Application Services overview, and aTODOcomment in the page marks the link to replace.DevOps links. The scenario pages and Disk Configuration link
install/planning-devops-cluster.html(anchorsharbor-on-globalanddevops-cluster), added by docs(install): add planning page for a dedicated DevOps cluster devops-docs#162.sites.yamlnow pinsdevopsto4.4. Every other DevOps link resolves onrelease-4.4of alauda/devops-docs: the site root,upgrade/index.html, andupgrade/migrating-tekton-v3-to-v4.html. The planning page exists onrelease-4.4only after Compliance service register #162 is backported, so these links do not open until then.Fixes an existing broken link in Upgrade the Builds Operator: the Tekton v3 to v4 migration guide moves from
4.0/upgrade/migrating-tekton-v3-to-v4.md, which does not exist on any DevOps release, toupgrade/migrating-tekton-v3-to-v4.html.Needs confirmation from the storage team:
ceph-mondaemon).External links use
ExternalSiteLink, so their versions followsites.yaml(logging-service 4.4, ai 2.8, devops 4.4). The ClickHouse and Kafka links are removed with the placement list.Needs confirmation from reviewers:
node-role.kubernetes.io/infrahas an empty value, so the steps tell users to leaveNODE_AFFINITY_VALUEempty. Please confirm that the quickstart script applies the node affinity in that case./var/cpaaschanges from "100 GB on the global control plane, 40 GB on other nodes" to "100 GB on control plane nodes, 50 GB on other nodes". The recommended/var/lib/kubeletis 200 GB onglobalcluster nodes and 100 GB on workload cluster nodes. The recommended/var/cpaasis 200 GB on log storage nodes, 100 GB on control plane nodes, and 50 GB on other nodes.Verification
yarn linton the whole tree: 0 errors, 0 warnings. A temporary page with a broken anchor in each changed directory was reported, then removed.